Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Heegner Divisors and Nonholomorphic
Modular Forms
JENS FUNKE$
Department of Mathematics, Rawles Hall, Indiana University, Bloomington,IN 47405, U.S.A. e-mail: [email protected]
(Received: 16 March 2001; accepted: 6 July 2001)
Abstract. We consider an embedded modular curve in a locally symmetric space M attached
to an orthogonal group of signature ðp; 2Þ and associate to it a nonholomorphic elliptic mod-ular form by integrating a certain theta function over the modular curve. We compute theFourier expansion and identify the generating series of the (suitably defined) intersection num-
bers of the Heegner divisors inM with the modular curve as the holomorphic part of the mod-ular form. This recovers and generalizes parts of work of Hirzebruch and Zagier.
Mathematics Subject Classi¢cations (2000). 11F11, 11F27, 11F30.
Key words. theta series, modular forms, Heegner divisors, intersection numbers.
1. Introduction
In their celebrated paper, Hirzebruch and Zagier [7] show that the intersection num-
bers of certain algebraic cycles on a Hilbert modular surface S occur as the Fourier
coefficients of holomorphic modular forms on the upper half plane H. They expli-
citly compute the intersection numbers of certain curves TN both in S and at the
resolution of the cusp singularities. By direct computation, they then show that these
numbers are Fourier coefficients of modular forms.
These results have inspired numerous other people to look at geodesic cycles in
other locally symmetric spaces and their relationship to modular forms. Here the case
SOð2; qÞwas of particular interest and was studied by Oda [19], Rallis and Schiffmann
[20], Kudla [9] and, recently, by Borcherds [1, 2] and Bruinier [3, 4].
Starting in the late 1970s and throughout the 1980s, Kudla and Millson (see, e.g.,
[14]) carried out an extensive program to explain the work of Hirzebruch–Zagier
from the point of view of Riemannian geometry and the theory of reductive dual
pairs and the theta correspondence. Under some restrictions, they vastly generalize
the results of [7] to orthogonal, unitary, and symplectic groups of arbitrary dimen-
sion and signature. Tong and Wang ([24]) ran a parallel program.
$Fellow of the European PostDoc Institute (EPDI) 99-01.
Compositio Mathematica 133: 289–321, 2002. 289# 2002 Kluwer Academic Publishers. Printed in the Netherlands.
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
Currently, Kudla, Rapoport and Yang undertake a major investigation of the
occurrence of arithmetic intersection numbers in certain moduli spaces as Fourier
coefficients of modular forms, see, e.g., [11].
This paper deals with the case of the real orthogonal group of signature ðp; 2Þ.
Before we state our results, we need to establish the basic notions of the paper. Let
ðVðQÞ; qÞ be a rational quadratic space of dimension p þ 2 and signature ðp; 2Þ and
let ð ; Þ be the associated nondegenerate symmetric bilinear form on VðQÞ. We have
ðv;wÞ ¼ qðv þ wÞ � qðvÞ � qðwÞ. We let G ¼ SpinðVðQÞÞ viewed as an algebraic group
over Q and write D ¼ GðRÞ=K for the associated symmetric space, where K is a max-
imal compact subgroup of GðRÞ. It is very well known that D is of Hermitian type of
complex dimension p; for example, for p ¼ 1, D ’ H, the upper half plane, and
D ’ H � H for p ¼ 2. We identify D with the space of two-dimensional subspaces
of VðRÞ on which the bilinear form ð ; Þ is negative definite:
D ’ fz VðRÞ : dim z ¼ 2 and ð ; Þjz < 0g: ð1:1Þ
Let L VðQÞ be an integral Z-lattice of full rank, i.e., L L#, the dual
lattice, and G be a congruence subgroup of G preserving L (in the main body of
the paper we will allow a congruence condition as well). We write M ¼ GnD for
the attached locally symmetric space of finite volume. We construct special cyclesin M as follows. Let U V be a positive definite subspace of V of any dimension
04 n4 p. We put
DU ¼ fz 2 D : z U?g; ð1:2Þ
so DU is a complex submanifold of the (same) orthogonal type Oðp � n; 2Þ and codi-
mension n. We can naturally identify SpinðU?Þ ’ GU, where GU is the pointwise sta-
bilizer of U in G. We put GU ¼ G \ GU and define the special cycle
CU ¼ GUnDU: ð1:3Þ
The map CU ! GnD defines an algebraic cycle in M and is actually an embedding if
one passes to a suitable subgroup G0 G of finite index (see [14]). For x 2 VðQÞn, we
denote by UðxÞ the subspace generated by x (of possibly lower dimension) and set
Dx ¼ DUðxÞ and Cx ¼ GxnDx. We will be mainly concerned with the case n ¼ 1, when
the special cycles are divisors. For N 2 N, G acts on LN ¼ fx 2 L : qðxÞ ¼ Ng with
finitely many orbits, and we define the composite cycle
CN ¼X
x2GnLN
Cx: ð1:4Þ
We call CN the Heegnerdivisor of discriminant N. For p ¼ 1, CN is the collection of
Heegner points of discriminant N in a modular (or Shimura) curve, while for p ¼ 2,
CN is a Hirzebruch-Zagier curve TN([7]) in a Hilbert modular surface (if the Q-rank
of G is 1).
Kudla and Millson explicitly construct (in much greater generality) a theta func-
tion yjðt;LÞ ¼P
x2L jðx; tÞ (t ¼ u þ iv 2 H) with values in the closed differential
290 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
ð1; 1Þ-forms of M attached to a certain Schwartz function j ¼ jV on VðRÞ (In Sec-
tion 2, we review this theta function in more detail). They then consider
Ijðt;CÞ ¼
ZC
yjðt;LÞ; ð1:5Þ
the integral of yjðt;LÞ over a compact curve C. Ijðt;CÞ turns out to be a holomorphicmodular form of weight ðp þ 2Þ=2, whose Nth Fourier coefficient is given as the
(cohomological) intersection number of C with the composite cycle CN. This gives
analogues of the original results of Hirzebruch–Zagier in a much more general set-
ting, but actually does not contain their work as they consider the intersection num-
bers of (in general) noncompact cycles. It therefore seems quite naturally to study the
theta integral (1.5) for possibly noncompact curves C. We study the theta integral
Ijðt;CÞ in the noncompact case, where we restrict our attention to the special curves
CU, with U positive definite of dimension p � 1, i.e. to embedded quotients of mod-
ular curves in M. This leads to considerable complications because in [14] the
assumption that C be compactly supported is quite essential and needed at several
places; for example, to guarantee the convergence of the integral (1.5), to show
the holomorphicity in t 2 H, and to verify the vanishing of the negative Fourier
coefficients.
In Section 3 we consider the case p ¼ 1 when M ’ GnH is itself a modular curve,
i.e., V is isotropic, and study the integral Ijðt;MÞ ¼RM yjðt;LÞ. Note that the cusps
of M correspond to the G-equivalence classes of isotropic lines ‘ in V. Our main
result is then that the generating series PðtÞ ¼P1
N¼0 degðCNÞqN of the degree of
Heegner points in M is the holomorphic part of a nonholomorphic modular form
of weight 3=2. Here
degðCNÞ ¼X
x2GnLN
1
jGxjfor N > 0;
while for N ¼ 0 we put degðC0Þ ¼ volðMÞ. More precisely:
THEOREM 1.1.ZM
yjðt;LÞ ¼ PðtÞ þv�1=2
4p
Xcusps ‘
Eð‘;L;GÞXN2Z
bð4pdk2‘N2vÞq�dk2‘N
2
is a nonholomorphic elliptic modular form of weight 3=2. Here Eð‘;L;GÞ denotes the
‘width’ of the cusp ‘ of G ðsee Def: 3.2Þ, d 2 N square-free, the discriminant of the
quadratic space V, k‘ the smallest k 2 N such that L�dk2;‘ ¼ fx 2 L�dk2 : x ? ‘g is
nonempty, and bðyÞ ¼ b3=2ðyÞ ¼R1
1 t�3=2e�ty dt.
For example, specializing to a certain lattice in the quadratic space of discriminant
1, we recover Zagier’s [25] well-known Eisenstein series F of weight 3=2 as a theta
integral. One has
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 291
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
F ðtÞ ¼X1N¼0
HðNÞe2pint þv�1=2
16p
X1N¼�1
bð4pN2vÞe�2piN2t; ð1:6Þ
where HðNÞ denotes the class number of positive definite binary quadratic forms of
discriminant �N2. From this perspective, we can consider Theorem 1.1 on one hand
as a special case of the Siegel–Weil formula expressing the theta integral as an Eisen-
stein series, and on the other hand as the generalization of Zagier’s function to arbi-
trary lattices of signature ð1; 2Þ.
COROLLARY 1.2.RM yjðt;LÞ � 4
P‘
ffiffiffid
pk‘Eð‘;L;GÞF ðdk2‘tÞ is a holomorphic
modular form. Hence, #ðGnLNÞ can be expressed in terms of class numbers and Fouriercoe⁄cients ofa holomorphic modular form.
We also would like to mention the results of [15], where it was shown that the gen-
erating series of the degree of certain 0-cycles in an arithmetic curve (namely, the
moduli scheme of elliptic curves with complex multiplication over the ring of integers
of an imaginary quadratic field) is the holomorphic part of a non-holomorphic mod-
ular form of weight 1. Here the negative Fourier coefficients involve the function
b1ðyÞ ¼R1
1 t�1e�tydt. The similarities of the situation over C and over number fields
are quite striking. We also compute the Mellin transform LðsÞ ofRM yjðt;LÞ. It turns
out to be closely related to certain Siegel zeta functions associated to (split) indefinite
quadratic forms of signature ð1; 2Þ which were previously studied by Shintani and
F. Sato (see [22, 21]). We have
THEOREM 1.3.
LðsÞ ¼ ð2pÞ�sGðsÞz2ðs;LÞ þ 2�1�sp�12�sG s � 1
2
� � 1sF 3
2; s; s þ 1;�1� �
z1ðs;LÞ;
where F ¼ 2F1 is the hypergeometric function, and the Siegel zeta functions are defined
by
z1ðs;LÞ ¼Xx2L
x? split
jqðxÞj�s and z2ðs;LÞ ¼Xx2L
qðxÞ>0
1
jGxjjqðxÞj�s:
The proof of Theorem 1:1 is long and occupies Section 4. After showing the con-
vergence ofRM yjðt;LÞ using a Poisson summation argument, we are reduced to cal-
culating the orbital integralsZGnD
Xg2GxnG
g�jðxÞ ð1:7Þ
for elliptic, hyperbolic, and parabolic stabilizer Gx. (In the parabolic case, Gx shall
mean here the stabilizer of the isotropic line generated by x, and one first has to
sum over all isotropic vectors before integrating.)
292 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
We treat the theta integral ð1:5Þ for general p and a modular curve C ¼ CU in
Section 5. The result is
THEOREM 1.4. LetU Vbe positive de¢nite ofdimension p � 1 so thatCU ’ GUnH.Assume for simplicity L ¼ L \ U þ L \ U?. Then
RCU
yjVðt;LÞ is nonholomorphic for
CU noncompact andZCU
yjVðt;LÞ ¼ yðt;L \ UÞ
ZCU
yjU?ðt;L \ U?Þ;
where yðt;L \ UÞ ¼P
x2L\U e2piqðxÞt is the standard theta series attached to the positive
definite lattice L \ U and the second integral is the one considered in Th. 1:4 for the
space U?, which has signature ð1; 2Þ. Moreover, the holomorphic Fourier coefficients
can be interpreted as the intersection numbers in ðthe interior of Þ M of the curve CU
and the divisor CN.
This generalizes parts of the results of Hirzebruch–Zagier, namely the ones con-
cerning the intersection numbers of the curves TN in the ‘interior’ of the Hilbert
modular surface. As an example for Theorem 1.2 we then explicitly derive these
parts. From there one can obtain the complete result of Hirzebruch–Zagier by
applying the holomorphic projection principle for modular forms to the theta
integral. This will then account for the contribution of the cusps to the intersec-
tion numbers (This is an idea of van der Geer and (independently) Zagier; see
[23].) However, when using this procedure, one still needs explicit formulas for
the intersection at the cusps. As this seems infeasible for the higher dimensional
case, a more conceptual approach for the cusps (similar to the treatment of the
‘interior’ presented here) is still needed. Our results should be closely related to
recent work of Borcherds [2] and Bruinier [3, 4]. They showed that the generating
seriesP1
N¼0 CNqN of Heegner divisors, when considered as elements in the so-
called ‘Heegner divisor class group’, is a holomorphic modular form of weight
ðp þ 2Þ=2 with values in this group in the sense that application of a linear form
on this Heegner divisor class group gives a scalar-valued holomorphic modular
form. One should expect that for p5 2, an extension of the methods used here
to the cusps should yield the results of [2, 3, 4] (while it seems that the methods
used there will not imply our results). For p ¼ 1, the results actually
are independent of each other, as taking the degree is the zero map on the
‘Heegner divisor class group’.
2. Work of Kudla and Millson
Kudla and Millson explicitly construct for orthogonal, unitary, and symplectic
groups of arbitrary signature the Poincare dual form of the cycles CU ([12, 13]).
We denote by Or;sðDÞ the space of smooth differentials forms of type ðr; sÞ on D.
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 293
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
THEOREM 2.1 ([12, 13]). For each n with 04 n4 p, there is a nonzero Schwartz form
jðnÞ 2 SðVðRÞnÞ � On;n
ðDÞ½ �G
ð2:1Þ
such that
ðiÞ djðnÞ ¼ 0; i.e., for each x 2 Vn, jðnÞðxÞ is a closed ðn; nÞ-form on D which is Gx-invariant: g�jðnÞðxÞ ¼ jðnÞðxÞ for g 2 Gx; the stabilizerof x inG.
ðiiÞ Denote by jþUðxÞ ¼ e�ptrðx;xÞ with x 2 Un and ðx; xÞi;j ¼ ðxi; xjÞ the standard Gaus-
sian on a positive de¢nite subspace U of VðRÞ. Then, under the pullbacki�U: On;n
ðDÞ ! On;nðDUÞ ofdi¡erential forms, we have
i�UjðnÞ ¼ jþ
U � jðnÞU? ;
where jðnÞU? is the nth Schwartz form for U? and GU.
ðiiiÞ Assume U ¼ UðxÞ for a linear independent n-frame in U.Then the Poincare¤ dual ofGUnDU is given by
epðx;xÞX
g2GUnG
g�jðnÞðxÞ
" #:
From now on we will restrict our attention to the case n ¼ 1. For simplicity we will
write j for the ð1; 1Þ-form jð1Þ.
Remark 2.2. We do not need the explicit general formula for the Schwartz form
right now; for an easily accessible construction see [9]. For signature ð1; 2Þ, we will
give the formula in the next section.
We denote by gSL2ðRÞ the two-fold cover of SL2ðRÞ. Recall that gSL2ðRÞ acts on the
Schwartz space SðVðRÞÞ via the Weil representation o associated to the additive
character t j�! expð2pitÞ, see for example [17]. Let K0 gSL2ðRÞ be the inverse image
of the standard maximal subgroup SOð2Þ in SL2ðRÞ.
PROPOSITION 2.3 ([12]). The Schwartz form j is an eigenfunction for K0 withrespect to theWeil representation i.e.,
oðk0Þj ¼ detðk0Þðpþ2Þ=2j: ð2:6Þ
We associate to j in the usual way a function on the upper half plane H. For
t ¼ u þ iv 2 H we put
g0t ¼
1 u0 1
� �v1=2 00 v�1=2
� �; ð2:7Þ
and define
jðt; xÞ ¼ v�ðpþ2Þ=4oðg0tÞjðxÞ: ð2:8Þ
294 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
Fix a congruence condition h 2 L# and assume that G fixes the coset L þ h under
the action on L#=L. Then the theta kernel
yjðtÞ ¼ yjðt;L; hÞ ¼X
x2hþL
jðt; xÞ 2 O1;1ðDÞ
Gð2:9Þ
defines a closed ð1; 1Þ-form inM ¼ GnD. By the usual theta machinery (Poisson sum-
mation) it is a nonholomorphic modular form for the congruence subgroup GðNÞ of
SL2ðZÞ (where N is the level of the lattice L), of weight ðp þ 2Þ=2 with values in
O1;1ðDÞ
G. Slightly more general than in the introduction we put
Lm ¼ fx 2 L þ h : qðxÞ ¼ mg for m 2 Q;
suppressing the dependence on h. For m > 0, we again obtain a composite divisor
Cm ¼P
x2GnLmCx in M. Let Z be a closed ðp � 1; p � 1Þ-form on M and assume that
Z is rapidly decreasing if M is noncompact. The main result of [14] (in much greater
generality) is that
Ijðt; ZÞ ¼
ZM
yjðtÞ ^ Z ð2:10Þ
is a holomorphic modular form whose Fourier coefficients are periods of Z over the
composite cycles Cm in M. If Z now represents the Poincare dual class of a compactcurve C in M, we obtain
Ijðt;CÞ ¼
ZC
yjðtÞ ¼ Ijðt; ZÞ; ð2:11Þ
and the Fourier coefficients are the (cohomological) intersection numbers of C with
Cm. In the following we consider (2.11) for modular curves C.
3. The Theta Integral Associated to SO ð1; 2Þ
3.1. PRELIMINARIES
Now assume dimV ¼ 3; hence V has signature ð1; 2Þ. Over R we fix an isomorphism
VðRÞ ’x1 x2x3 �x1
� �2 M2ðRÞ
ð3:1Þ
such that
qðXÞ ¼ detðXÞ ¼ �x21 � x2x3 and ðX;YÞ ¼ �trðXYÞ:
So we can view VðRÞ as the trace zero part B0ðRÞ of the indefinite quaternion algebra
BðRÞ ¼ M2ðRÞ over R. We have G ¼ SpinðVÞ ¼ SL2 and the action on B0 is the con-
jugation:
g � X :¼ gXg�1 ð3:2Þ
for X 2 B0 and g 2 G.
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 295
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
Notation. In this section, we will write z ¼ x þ iy for an element in D ’ H (the
orthogonal variable) and t ¼ u þ iv 2 H for the symplectic variable. The upper case
letters X and Y we reserve for vectors in VðRÞ with coefficients xi and yi.
Here it is more convenient to consider the symmetric space D ’ H not as the space
of negative two-planes in VðRÞ but rather as the space of positive lines. We give the
following identification with the upper half plane. Picking as base point of D the line
z0 spanned by�0 1�1 0
�, we note that K ¼ SOð2Þ is its stabilizer in GðRÞ, and we have
the isomorphism:
H ’ GðRÞ=K ! D ð3:3Þ
with
z j�! gK j�! g � z0 ¼: ‘ðzÞ ð3:4Þ
(where g 2 GðRÞ such that gi ¼ z; the action is the usual linear fractional transforma-
tion). We find that ‘ðzÞ is generated by
XðzÞ :¼ y�1 � 12 ðz þ �zÞ z �z�1 1
2 ðz þ �zÞ
� �; ð3:5Þ
where z ¼ x þ iy. Note qðXðzÞÞ ¼ 1. Moreover, per construction
g � XðzÞ ¼ XðgzÞ: ð3:6Þ
For X ¼�x1 x2x3 �x1
�2 B0ðRÞ we compute
ðX;XðzÞÞ ¼ �y�1ðx3z �z � x1ðz þ �zÞ � x2Þ
¼ �y�1ðx3ðx2 þ y2Þ � 2x1x � x2Þ
¼ðx3x � x1Þ
2þ qðXÞ
�x3y� x3y; ð3:7Þ
when x3 6¼ 0.
The minimal majorant of ð ; Þ associated to z 2 D is given by
ðX;XÞz ¼ðX;XÞ; if X 2 ‘ðzÞ;
�ðX;XÞ; if X 2 ‘ðzÞ?:
ð3:8Þ
A little calculation shows
ðX;XÞz ¼ ðX;XðzÞÞ2 � ðX;XÞ
¼ðx3x
2 � x2 � 2x1xÞ2
y2þ ðx3yÞ
2þ 2ðx3x � x1Þ
2: ð3:9Þ
PROPOSITION 3.1 ([10]). The Schwartz function j ¼ jð1Þ on VðRÞ valued in theð1; 1Þ-forms onD is explicitly given by
jðX; zÞ ¼
�ðX;XðzÞÞ2 �
1
2p
�e�pðX;XÞz o: ð3:10Þ
296 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
Here
o ¼dx ^ dy
y2¼
i
2
dz ^ d �z
y2; ð3:11Þ
the standard G-invariant ð1; 1Þ-form on D ’ H.
We will write jðXÞ for its value at X. Then
g�jðXÞ ¼ jðg�1XÞ; ð3:12Þ
i.e., jðg � X; gzÞ ¼ jðX; zÞ for g 2 GðRÞ, follows from (3.6); while Proposition 2.3
becomes an exercise using the explicit formulae of the Weil representation and redu-
ces to j ¼ �j. Finally, we define
j0ðX; zÞ ¼ epðX;XÞjðX; zÞ ð3:13Þ
¼ ðX;XðzÞÞ2 �1
2p
� �e�pðX;XðzÞÞ2þ2pðX;XÞ o ð3:14Þ
and
jðt;XÞ ¼
�vðX;XðzÞÞ2 �
1
2p
�epiðX;XÞz;t o; ð3:15Þ
where
ðX;XÞz;t ¼ uðX;XÞ þ ivðX;XÞz:
¼ �tðX;XÞ þ ivðX;XðzÞÞ2: ð3:16Þ
3.2. THE THETA INTEGRAL
We put
IjðtÞ :¼ZGnD
yðt;L; hÞ ¼
ZGnD
XX2Lþh
jðt;XÞ: ð3:17Þ
Thus, in the notation of the previous section, IjðtÞ ¼ yjðt; ZÞ where Z is the constant
function 1. In particular, Z is not rapidly decreasing. Alternatively we can interpret
(3.21) as the integral Ijðt;CUÞ in the case of signature ð1; 2Þ with U ¼ 0. We assume
the convergence of (3.17) for the moment. Then it is again clear that IðtÞ defines a (in
general nonholomorphic) modular form on the upper half plane of weight 3=2. For
X 2 Lm, we have
jðt;XÞ ¼ qmj0ðffiffiffiv
pXÞ; ð3:18Þ
where qm ¼ e2pimt, as usual. Define
ymðtÞ ¼XX2Lm
jðt;XÞ and y0mðvÞ ¼XX2Lm
j0ðffiffiffiv
pXÞ: ð3:19Þ
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 297
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
We then – yet formally – have
IjðtÞ ¼
ZGnD
Xm2Q
ymðtÞ ¼Xm2Q
ZGnD
y0mðvÞ
� �qm; ð3:20Þ
which is the Fourier expansion of IjðtÞ. Since GnLm is finite for m 6¼ 0, we can
simplify the inner integral in (3.20) in that case:ZGnD
y0mðvÞ ¼
ZGnD
XX2GnLm
Xg2GXnG
j0ðg�1ffiffiffiv
pX; zÞ
¼X
X2GnLm
ZGnD
Xg2GXnG
g�j0ðffiffiffiv
pX; zÞ: ð3:21Þ
For X ¼ 0, we simply have jðt;XÞ ¼ �ð1=2pÞo and thereforeZGnD
jðt; 0Þ ¼ mðGnDÞ; ð3:22Þ
the hyperbolic volume of GnD, normalized such that mðSL2ðZÞnHÞ ¼ � 16. In particu-
lar, we see that with this normalization (see, e.g., [18])
mðGnDÞ 2 Q: ð3:23Þ
If V is isotropic over Q; we can pick the isomorphism (3.1) such that
VðQÞ ’
ffiffiffid
px1 x2
x3 �ffiffiffid
px1
� �: xi 2 Q
¼: B0ðd;QÞ ð3:24Þ
as quadratic Q-vector spaces, where d is a square-free positive integer, the discrimi-
nant of the quadratic space V. The set of all isotropic lines in V corresponds to the
cusps of GðQÞ and G acts on them with finitely many orbits. For the constant term in
the Fourier expansion of IjðtÞ, we therefore have to proceed differently than in the
case m 6¼ 0: Let ‘1; . . . ; ‘t be a set of G-representatives of isotropic lines and pick
Xi 2 ‘i primitive in L. Define dð‘iÞ ¼ dð‘i;L; h;GÞ ¼ 1 if ‘i intersects the coset
L þ h and dð‘iÞ ¼ 0 otherwise. Hence, ‘i \ ðL þ hÞ ¼ ZXi þ hi for some hi 2 ‘i if
dð‘iÞ ¼ 1. Denote by Gi the stabilizer of ‘i in G. We then haveZGnD
XX2L0X 6¼0
j0ðffiffiffiv
pX; zÞ ¼
ZGnD
Xti¼1
XX2Gð‘i\ðLþhÞÞ
X6¼0
j0ðffiffiffiv
pX; zÞ
¼
ZGnD
Xti¼1
XX2‘i\ðLþhÞ
X 6¼0
Xg2GinG
g�j0ðffiffiffiv
pX; zÞ
¼Xti¼1
dð‘iÞZGnD
Xg2GinG
X1k¼�1
0
g�j0ðffiffiffiv
pðkXi þ hiÞ; zÞ: ð3:25Þ
HereP0 indicates that we omit k ¼ 0 in the sum in the case of the trivial coset. We
need to discuss the notion of the width of a cusp for our purposes. For Y 2 V
298 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
isotropic (usually primitive in L), pick g 2 GðRÞ such that gY ¼ bX0 with
X0 ¼�0 10 0
�and b 2 R�. Put G0 ¼ gGg�1. Hence, G0
X0¼ gGYg
�1 is equal to���1 ka0 1
�: k 2 Z
�(if �I 2 G) for some a 2 Rþ. Now we could call a the width
of the cusp ‘, however this is not well defined, since it depends on the choice of
g 2 GðRÞ and, hence, on b. However, one easily checks that the ratio a=jbj is constantand only depends on Y and G.
DEFINITION 3.2. (Width of a cusp). (i) For Y isotropic, we define EðY;GÞ ¼ a=jbj;where a and b are the quantities above. We call the pair ðY;GÞ a cusp and the number
EðY;GÞ its width.
(ii) For a lattice L V, h 2 L#=L and G GðLÞ, we define (with the above nota-
tion) the total width by EðL; h;GÞ ¼Pt
i¼1 dð‘iÞEðXi;GÞ:
Remark 3.3. (i) If G is a congruence subgroup of SL2ðZÞ, then there is (up to sign)
a unique g 2 SL2ðZÞ such that gY ¼ bX0 ¼ b�0 1
0 0
�. Then both numbers a and b are
intrinsic: a is simply the usual width of a cusp of a congruence subgroup of SL2ðZÞ
while b can be interpreted as the volume of the fundamental domain of RX0=ZbX0
with respect to the Lebesgue measure on RX0.
(ii) We can always arrange jbj ¼ 1. Then we can interpret a ¼ EðY;GÞ as the
volume of the corresponding component of the Borel–Serre boundary of M.
(iii) It does indeed happen that for a fixed space V we can find two lattices with the
same stabilizer G such that the ’cusp 1’ has different width. Consider the lattices
b ac �b
� : a; b; c 2 Z
n oand b 2a
2c �b
� : a; b; c 2 Z
n o:
Both have stabilizer G ¼ SL2ðZÞ, hence a ¼ 1, but b ¼ 1 and 2, respectively. See also
Example 3.9.
We are now ready for the main result:
THEOREM 3.4.With the above notationwe have
ðiÞ yjðtÞ 2 L1ðGnDÞ;
ðiiÞ y0mðvÞ 2 L1ðGnDÞ;
forallm 2 QandZGnD
yjðtÞ ¼Xm2Q
ZGnD
y0mðvÞ
� �qm:
For the Fourier coefficients; we get
ðiiiÞ form > 0:ZGnD
y0mðvÞ ¼X
X2GnLm
1
jGXj;
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 299
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
ðivÞ form ¼ 0:ZGnD
y0mðvÞ ¼ mðGnDÞ þ1
2pv�1=2EðL; h;GÞ;
where the volume term only occurs for h 2 L;(v) form < 0:Z
GnD
y0mðvÞ ¼1
4pffiffiffiffiffiffiffiffi�m
p v�1=2X
X2GnLmX? isotropic
Z 1
1
t�3=2e4pvmtdt
� �:
Note that for m < 0 we haveZ 1
1
t�3=2e4pmtdt4Oðe4pmÞ ð3:26Þ
so that the negative part of the Fourier expansion has the same convergence
behavior as the positive part. The next section will deal with the proof of the
Theorem 3.4.
First we discuss the convergence of IjðtÞ. Then we turn our attention to the com-
putation of the individual integrals which define the Fourier coefficients via the ana-
lysis (3.19) to (3.25). For m > 0, the composite 0-cycle Cm ¼P
X2GnLmCX is the
collection of Heegner points of discriminant m, and we define its degree by
degðCmÞ ¼X
X2GnLm
1
jGXj: ð3:27Þ
We also put
degðC0Þ ¼mðGnDÞ; if 0 2 L þ h,0; else.
ð3:28Þ
In any case we have degðCmÞ 2 Q. We define the generating series of the degree of the
Heegner points by
PðtÞ ¼Xm5 0
degðCmÞqm: ð3:29Þ
Then as a corollary we obtain the following generalization of the results of Kudla
and Millson to the noncompact case:
THEOREM 3.5.LetVbe a quadratic space over Qof signature ð1; 2Þ.
ðiÞ ð½14�Þ If V is anisotropic, i.e., GnD compact, then the generating function PðtÞ is aholomorphic modular form of weight 3=2 for a suitable congruence subgroup ofSL2ðZÞ.
300 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
ðiiÞ If V is isotropic, i.e., GnD noncompact, then PðtÞ is the holomorphic part of the non-holomorphic modular form of weight 3=2given byXm5 0
degðCmÞqm þv�1=2
2pEðL; h;GÞ þ
Xm>0
v�1=2
4pffiffiffiffim
pX
X2GnL�mX? isotropic
bð4pvmÞ q�m; ð3:30Þ
where we set; following Zagier ½25� ðup to a factorÞ, bðsÞ ¼R1
1 t�3=2e�stdt.
The following lemma characterizes the values for which the negative Fourier coef-
ficients in the previous theorem are nonzero.
LEMMA 3.6. For X 2 VðQÞ ’ B0ðd;QÞ with qðXÞ < 0 the following two statements areequivalent:
ð1Þ X? is split over Q.ð2Þ qðXÞ 2 �d Q
�ð Þ
2.
Proof. If qðXÞ ¼ �dm2 < 0 with m 2 Q, then by Witt’s Theorem we can map
X 7!�m
ffiffid
p0
0 �mffiffid
p�
2 B0ðd;QÞ; hence X? is split. Conversely, if X? is split, we can
assume X ?�0 1
0 0
�(again by Witt’s Theorem moving an isotropic vector orthogonal
to X to�0 1
0 0
�. Thus X ¼
�m
ffiffid
p�
0 �mffiffid
p�for some m 2 Q. &
Let X 2 L�dm2 . Then X is orthogonal to two cusps. However, giving an orientation
to QX we can distinguish between these cusps (if they are not equivalent), since
switching the cusps by an element in G \ SOðX?Þ switches X to �X. For a fixed cusp
‘i, we write
L�dm2;i;þ ¼ fX 2 L�dm2 : X ? ‘i; X pos orient:g
and note that Gi acts on this set. We have
LEMMA 3.7.
#GnL�dm2 ¼Xti¼1
#GinL�dm2;i;þ
and
#GinL�dm2;i;þ ¼ 2mffiffiffid
pEðXi;GÞ;
if L�dm2;i;þ is not empty.
Proof. The first assertion is clear. For the second, take X 2 L�dm2 , say
X ¼�m
ffiffid
p0
0 �mffiffid
p�. So X is orthogonal to the cusps 0 and 1. We distinguish them by
requiring that the left upper left entry of X is positive, since switching the cusps by�0 1
�1 0
�maps X to �X. By our conventions about the cusps we can assume that�
0 b0 0
�is primitive in L with stabilizer
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 301
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
G1ðaÞ ¼ Tak ¼ �1 ak0 1
� �: k 2 Z
in G ðif � I 2 GÞ:
Now
Tak � X ¼m
ffiffiffid
p�2m
ffiffiffid
pak
0 �mffiffiffid
p
� �; ð3:31Þ
hence
X � Tak � X ¼0 �2m
ffiffiffid
pak
0 0
� �2 L: ð3:32Þ
Thus 2mffiffiffid
pa 2 bZ. The assertion now follows from the observation that we have
showed that we have 2mffiffiffid
pE equivalence classes of vectors in L�dm2;i;þ. &
Lemmata 3.6 and 3.7 enable us to rewrite Theorem 3.5(ii):
THEOREM 3.8. With the above notationwe have
Ijðt;L; hÞ ¼Xm�0
degðCmÞqm þv�
12
2p
Xcusps ‘i
EðXi;GÞ�
� dð‘iÞ þXm2Qþ
L�dm2;i;þ6¼;
bð4pdm2vÞq�dm2
0BBBBBB@
1CCCCCCA:
This implies Theorem 1.1 as follows: For the constant Fourier coefficient note
that for the trivial coset we have dð‘iÞ ¼ 1 and bð0Þ ¼ 2. For the negative coeffi-
cients, take a nonisotropic vector X in ‘?i , primitive in L. We see qðXÞ ¼ �dk2‘i
for some k‘i 2 N. Now all other vectors in the sum for the cusp ‘i are integral mul-
tiples of X.
EXAMPLE 3.9 (Zagier’s Eisenstein series of weight 3=2). Let V be the space of trace
0 elements of the (indefinite) split quaternion algebra M2ðQÞ over Q; i.e.,
V ’ B0ð1;QÞ. (i). We first consider the isotropic lattice
L ¼b 2a2c �b
� �: a; b; c 2 Z
; ð3:33Þ
so qða; b; cÞ ¼ �b2 � 4ac. L has level 4. As mentioned above, we have G ¼ GðLÞ ¼
SL2ðZÞ and one isomorphism class of cusps. The stabilizer of ‘0 ¼ Q�0 2
0 0
�is
G1 ¼ f��1 n
0 1
�: n 2 Zg. Note that the width of this cusp (in the above sense) is
E ¼ 12! By Lemma 3.7 G acts on L�n2 with n orbits, and these X are precisely the
302 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
vectors such that X? is split over Q. For N > 0 and X ¼�
b 2a
2c �b
�such that
qðXÞ ¼ N, observe that the assignment
X j�!�2c bb 2a
� �ð3:34Þ
defines (for a > 0) a binary positive definite quadratic form of discriminant �N. One
easily deduces that
# GnLNð Þ ¼ 2HðNÞ; ð3:35Þ
where HðNÞ denotes the class number of binary positive definite integral quadratic
forms of discriminant �N; we count the classes with nontrivial stabilizer with multi-
plicities 1=2 and 1=3, respectively. The elements of the form k�
0 2
�2 0
�(corresponding
to i 2 H) are fixed by�
0 1
�1 0
�and the stabilizer of the elements of the form k
�1 2
�2 1
�(corresponding to ð1 þ i
ffiffiffi3
pÞ=2) is generated by
�1 �1
1 0
�. We set Hð0Þ ¼ � 1
12 ¼12 mðGnDÞ. Applying Theorem 1.1 we obtain
1
2Ijðt;LÞ ¼ F ðtÞ ¼
X1N¼0
HðNÞqN þv�1=2
16p
Xn2Z
bð4pn2vÞ q�n2 ; ð3:36Þ
Zagier’s well known nonholomorphic Eisenstein series of weight 3=2 and level 4,
see [25,7]. (ii). This example we will need later to deduce a case of the results of
Hirzebruch and Zagier. We almost repeat (i) considering the trace zero elements
in the maximal order M2ðZÞ in M2ðQÞ:
K ¼b ac �b
� �: a; b; c 2 Z
; ð3:37Þ
so qða; b; cÞ ¼ �b2 � ac. We set X1 ¼�1 0
0 �1
�and define (the coset)
Kj ¼j
2X1 þ K ð3:38Þ
for j ¼ 0; 1. Again we put G ¼ SL2ðZÞ ¼ GðKjÞ. This time we have E ¼ 1 and d ¼ 0
for j ¼ 1. As above we see that G acts on the vectors of length �ðn þ ðj=2ÞÞ2 with
2n þ j orbits. For N > 0 and X ¼�bþj=2 a
c �b�j=2
�such that qðXÞ ¼ N �
j4 we assign
X j�!�2c 2b þ j2b þ j 2a
� �; ð3:39Þ
which defines a positive form of discriminant �4N þ j! We obtain
#ðGnðKjÞN�j4Þ ¼ 2Hð4N � jÞ; ð3:40Þ
hence
1
2Ijðt;KjÞ ¼
X1N¼0
Hð4N � jÞqN�j=4 þv�1=2
8p
Xn2Z
b 4p n þj
2
� �2
v
!q�ðnþ j
2Þ2
: ð3:41Þ
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 303
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
The example now implies Corollary 1.2.
As a corollary to the example and as an illustration of Corollary 1.2 we obtain the
famous relationship between the representation numbers of integers as a sum of
three squares and the class numbers of binary quadratic forms:
COROLLARY 3.10. Denote by r3ðNÞ the representation number of N as a sum of
three squares and write WðtÞ ¼P
n2Z e2pin2t for the classical Jacobi theta series. Then
WðtÞð Þ3¼ 12
1
2Ijðt;K0Þ � Ijðt;LÞ
� �;
i.e.,
r3ðNÞ ¼ 12 Hð4NÞ � 2HðNÞð Þ
for all integers N.
Proof. The space of holomorphic modular forms for G0ð4Þ of weight 3=2 of
Nebentypus is one-dimensional, spanned by WðtÞð Þ3. &
We now compute the Mellin transform LðsÞ of IjðtÞ, hence giving the proof of
Theorem 1.3: We write
LðsÞ ¼ LþðsÞ þ L�ðsÞ ð3:42Þ
with
L�ðsÞ ¼
Z 1
0
XX2Lþh�qðXÞ>0
jðiv;XÞvsdv
v: ð3:43Þ
For the positive part, we obtain in the standard fashion via Theorem 3.4 (iii)
LþðsÞ ¼ ð2pÞ�sGðsÞX
X2LþhqðXÞ>0
1
jGxjqðXÞ
�s¼ ð2pÞ�sGðsÞz2ðs;L; hÞ: ð3:44Þ
For the negative part, we have via Theorem 3.4 (v)
L�ðsÞ ¼
Z 1
0
XX2LþhX? split
v�12
4pffiffiffiffiffiffiffiffiffiffiffiffijqðXÞj
p Z 1
1
u�32e�4pvjqðXÞjudu
� �e2pjqðXÞjvvs
dv
vð3:45Þ
¼X
X2LþhX? split
v�12
4pffiffiffiffiffiffiffiffiffiffiffiffijqðXÞj
p Z 1
1
Z 1
0
e�2pjqðXÞjð2u�1Þvvs�12dv
v
� �u�3
2du ð3:46Þ
¼X
X2LþhX? split
v�12
4pffiffiffiffiffiffiffiffiffiffiffiffijqðXÞj
p G s � 12
� � Z 1
1
ð2pjqðXÞjð2u � 1ÞÞ12�su�3
2du ð3:47Þ
¼ 2�32�sp�1
2�sG s � 12
� �z1ðs;L; hÞ
Z 1
1
ð2u � 1Þ�12�s
u3=2du: ð3:48Þ
304 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
The integral in the last line is equal to
ffiffiffi2
pZ 1
0
ws�1
ðw þ 1Þ3=2¼
ffiffiffi2
p 1
sF 3
2
� �; s; s þ 1;�1Þ;
see [16]. This concludes the proof of Theorem 1.3.
4. Proof of Theorem 3.4
4.1. CONVERGENCE OF THE THETA INTEGRAL
PROPOSITION 4.1 (Theorem 3.4 (i)). yjðt;L; hÞ 2 L1ðGnDÞ:
Proof. If GnD is compact, i.e., V ¼ VðQÞ is anisotropic, then there is nothing to
show; the existence of the integral is immediate. On the other hand, if V is isotropic
then we can choose the isomorphism of VðRÞ ’ B0ðRÞ such that VðQÞ ’ B0ðd;QÞ
and X0 :¼�0 1
0 0
�primitive in L. We then can find r 2 Q such that
L0 :¼ rB0ðd;ZÞ L. With this notation we see
yjðt;L; hÞ ¼X
h0�hðLÞmodL0
yjðt;L0; h0Þ: ð4:1Þ
So it is sufficient to show that each yjðt;L0; h0Þ 2 L1ðGnDÞ. Picking a fundamental
domain for GnD we observe that via (3.12) it suffices to show that yj is rapidly
decreasing as y ! 1. For X 2 B0ðd;QÞ, we have
jðt;XÞ ¼
�vðX;XðzÞÞ2 �
1
2p
�e�pvðX;XðzÞÞ2e2pi�tqðXÞo
¼v
y2ðx3z �z � 2
ffiffiffid
px1x � x2Þ
2�
1
2p
� ��
� exp �pv
y2ðx3z �z � 2
ffiffiffid
px1x � x2Þ
2
� ��
� exp ��tðdx21 þ x2x3Þ� �
o: ð4:2Þ
Write x2 ¼ x02 þ h0
2 and let x02 run over rZ. We will apply Poisson summation to the
sum on x02. So consider in the above expression the coefficient of o as a function
f of x02. For the Fourier transform of f, we see by changing variables to
�t ¼
ffiffiffiv
p
yðx3z �z � 2
ffiffiffid
px1x � x0
2 � h02Þ;
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 305
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
fðwÞ ¼
Z 1
�1
fðx02Þ exp ð�2pix0
2wÞdx02
¼ ðyffiffiffiv
p�1 exp
���tdx21
�expð�½w þ x3 �t�½x3z �z � 2
ffiffiffid
px1x � h0
2�Þ�
�
Z 1
�1
t2 �1
2p
� �expð�pt2Þ exp �t
yffiffiffiv
p ðw þ x3 �tÞ� �
dt
¼ �ð�1Þyffiffiffiv
p ðw þ x3 �tÞ� �2
exp �pyffiffiffiv
p ðw þ x3 �tÞ� �2
!; ð4:3Þ
since the Fourier transform of ðt2 � ð1=2pÞÞ expð�pt2Þ is �t2 exp ð�pt2Þ. We obtain:
yðt;L0; h0ÞðzÞ ¼ �r�1 y
v3=2
Xw2r�1Z
x12h01þrZ
x32h03þrZ
ðw þ x3 �tÞ2 expð��tdx21Þ�
� expð�½w þ x3 �t�½x3z �z � 2ffiffiffid
px1x � h0
2�Þ exp �py2
vðw þ x3 �tÞ
2
� �o:
All terms of this sum are exponentially decreasing as y 7!1 except those for which
w ¼ x3 ¼ 0. But the coefficient w þ x3 �tð Þ2 vanishes for such terms. So jyðt;L0; h0Þj is
integrable and yðt;L; hÞ is real analytic in t. &
4.2. COMPUTATION OF THE INDIVIDUAL TERMS
We will compute for G an arbitrary Fuchsian subgroup of SL2ðRÞ the integralsZGnD
Xg2GXnG
g�j0ðX; zÞ ð4:4Þ
with any nonisotropic X 2 VðRÞ and GX the stabilizer of X in G, andZGnD
XX2GðZYþhYÞ
X6¼0
j0ðX; zÞ; ð4:5Þ
where Y 2 VðRÞ isotropic such that QY ¼ ‘ is a cusp of G, i.e., GY is nontrivial, and
hY 2 ‘. Note that via (3.21) and (3.25) the calculation of these integrals yields the
Fourier expansion of IjðtÞ (Theorem 3.4).
There are several cases to consider:
(A) qðXÞ > 0;
(B1) qðXÞ < 0, GX nontrivial, infinite cyclic,
(B2) qðXÞ < 0, GX trivial,
(C) The integral (4.5).
According to the stabilizer GX of X (G‘ of the isotropic line ‘) we call (A) the
elliptic, (B) the hyperbolic and (C) the parabolic case. Note that the cases are closely
related to each other:
306 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
LEMMA 4.2. Let qðXÞ < 0 forX 2 VðQÞ, soX? has signature ð1; 1Þ.ThenGX is trivial ifX? splits over Q. Conversely, if X? is nonsplit, i.e., anisotropic over Q, then GX is in¢nitecyclic.
Proof. Indeed, we can identify GXðRÞ with SpinðX?Þ ’ R� which acts on the two
isotropic lines of X? as homotheties. Hence, GX ¼ GðX?Þ being a discrete subgroup
of R� is either trivial or infinite cyclic. But for a possible isotropic line ‘ over Q, i.e.,
a cusp for G, we would also have a discrete unipotent subgroup in G stabilizing ‘.
This together with GðX?Þ has an accumulation point, see [18], p. 16. On the other
hand, an indefinite anisotropic binary quadratic form over Q corresponds to the
norm form of a real quadratic field K over Q, and the units O�K act as isometries and
are infinite cyclic (up to torsion). &
So the cases (B2) and (C) occur together. Moreover, they occur if and only if GnD
is noncompact. These two cases create the complications when extending the results
of Kudla and Millson to the noncompact case.
4.2.1(A). The Elliptic Case
Let X 2 VðRÞ such that qðXÞ > 0, hence RX 2 D. Therefore the stabilizer GX of X in
GðRÞ ¼ SL2ðRÞ is conjugate to SO2ðRÞ, and is in particular compact. Since GX is a
discrete subgroup, it is finite cyclic.
PROPOSITION 4.3 (Theorem 3.4 (iii)). Let qðXÞ > 0.ThenXg2GXnG
jg�jðX; zÞj 2 L1ðGnDÞ;
unfolding in ð4:4Þ is allowed, and we haveZGnD
Xg2G
g�j0ðX; zÞ ¼
ZD
j0ðX; zÞ ¼ 1: ð4:6Þ
Remark 4.4. This result is certainly already contained in [12,13]. In fact, it is one of
the corner stones of the theory. However, for the convenience of the reader we give a
brief sketch of the proof.
Proof. Write X ¼�x1 x2x3 �x1
�. Since qðXÞ > 0, we have x3 6¼ 0. Hence by the explicit
formulae for j we get
j0ðX; zÞ ¼ e2pðX;XÞ ðx3x � x1Þ2þ qðXÞ
�x3y� x3y
� �2
�1
2p
" #�
� e�p
ðx3x�x1 Þ2þqðXÞ
�x3y�x3y
� 2
dxdy
y2: ð4:7Þ
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 307
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
The integrand is rapidly decreasing for x ! �1 and y ! 0;1. Note that here one
needs qðXÞ > 0. Hence unfolding is allowed and therefore j0ðX; zÞ 2 L1ðDÞ. The
proof ofRD j0ðX; zÞ ¼ 1 can be found in [13], p. 301–302. Note that the Schwartz
function considered there differs from ours by a factor of 1=2. &
4.2.2(B). The Hyperbolic Case
So let X 2 VðRÞ such that qðXÞ < 0, say qðXÞ ¼ �d with d 2 Rþ. The stabilizer of
X1ðdÞ :¼
ffiffiffid
p0
0 �ffiffiffid
p
� �in GðRÞ ¼ SL2ðRÞ
is
GX1ðdÞðRÞ ¼a 00 a�1
� �: a 2 R�
:
So GX is conjugate to a discrete subgroup G0X of GX1ðdÞ, hence either infinite cyclic or
trivial.
Case (B1). With the above notation assume that GX is infinite cyclic, say
g � X ¼ X1ðdÞ and gGXg�1 ¼ G0
X ¼� E 0
0 E�1
�with some E > 1 (can assume E > 0,
since �I acts trivially on D).
PROPOSITION 4.5 (Theorem 3.4(v)). Let X 2 VðRÞ such that qðXÞ < 0 and GX
in¢nite cyclic.ThenXg2GXnG
jg�j0ðX; zÞj 2 L1ðGnDÞ;
unfolding in ð4:4Þ is valid andZGnD
Xg2GXnG
g�j0ðX; zÞ ¼
ZGXnD
j0ðX; zÞ ¼ 0: ð4:8Þ
Remark 4.6. This orbital integral also appears in the compact quotient case. Since
in that case all negative Fourier coefficients vanish, the above proposition also fol-
lows from the work of Kudla and Millson. However, as they only sketch the
argument in this particular case ([14], p. 138), it seems desirable to give a direct
proof.
Proof. We haveZGnD
Xg2GXnG
g�j0ðX; zÞ ¼
ZgGg�1nD
Xg2GXnG
j0ðg � X; ggg�1zÞ
¼
ZG0nD
Xg2G0
XnG0
g�j0ðX1ðdÞ; zÞ ð4:9Þ
308 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
where G0 :¼ gGg�1. We will now show the validity of the unfolding which then
proves the existence of the original integral as well.ZG0nD
Xg2G0
XnG0
g�j0ðX1ðdÞ; zÞ ¼
ZG0XnD
j0ðX1ðdÞ; zÞ
¼ e�4pdZG0XnD
4dx2
y2�
1
2p
� �e�4pd x2
y2dxdy
y2: ð4:10Þ
� E 0
0 E�1
�acts on z 2 D as z ! E2z. So a fundamental domain F of G0
XnD is the
domain bounded by the semi arcs jzj ¼ 1 and jzj ¼ E2 > 1 in the upper half plane:
F ¼ z 2 D : 14 jzj < E2� �
: ð4:11Þ
But in this domain j0ðX1ðdÞ; zÞ is clearly rapidly decreasing as z approaches the
boundary of D. So all considered integrals actually exist and unfolding is allowed.
Changing to polar coordinates we computeZF
4dx2
y2�
1
2p
� �e�4pdx
2
y2dxdy
y2
¼
Z E2
1
Z p
0
4d cot2ðyÞ �1
2p
� �e�4pd cot2ðyÞ 1
rcsc2ðyÞdydr
¼ logðE2ÞZ 1
�1
4pdt2 �1
2p
� �e�4pdt2dt ðt ¼ cot yÞ
¼ 0; ð4:12Þ
since Zt2 �
1
2p
� �e�pt2dt ¼
1
2pte�pt2 þ C: ð4:13Þ
&Case (B2).
PROPOSITION 4.7 (Theorem 3.4(v)). Let qðXÞ be negative and assume that GX istrivial.ThenX
g2G
g�j0ðX; zÞ 2 L1ðGnDÞ ð4:14Þ
and ZGnD
Xg2G
g�j0ðX; zÞ ¼1
4pffiffiffiffiffiffiffiffiffiffiffiffijqðXÞj
p Z 1
1
t�3=2e4pqðXÞtdt: ð4:15Þ
Proof. Let qðXÞ ¼ �d and suppose that GX is trivial. Recall that we chose g 2 GðRÞ
such that
g � X ¼ X1ðdÞ ¼
ffiffiffid
p0
0 �ffiffiffid
p
� �and gGg�1 ¼ G0:
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 309
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
As above we haveZGnD
Xg2G
g�j0ðX; zÞ ¼
ZG0nD
Xg2G0
g�j0ðX1ðdÞ; zÞ: ð4:16Þ
Unfolding in this situation will be not possible since
j0ðX1ðdÞ; zÞ ¼ e�4pd 4dx2
y2�
1
2p
� �e�4pdx
2
y2dxdy
y2ð4:17Þ
is not integrable over D. So we have to proceed more carefully. GX is trivial. Hence
X?, having signature ð1; 1Þ, is split. Therefore the stabilizers in G of the isotropic lines
in X? are nontrivial by Lemma 4.2. By conjugation we consider the isotropic lines in
X1ðdÞ?. They are generated by
X0 ¼0 10 0
� �and fX0 ¼
0 0�1 0
� �:
We put J ¼�
0 1
�1 0
�. Note that J switches the isotropic lines (i.e. the cusps) orthogonal
to X1ðdÞ: JX0 ¼ fX0. Hence, G0X0
¼ hTai with Ta ¼�1 a0 1
�for some a 2 Rþ and
G0eX0
¼ hJTbJ�1i for some b 2 Rþ. We write G00
X0¼ hTbi. Note that a and b are not
intrinsic to the situation since they depend on the choice of g such that
g � X ¼ X1ðdÞ. Define a subset G of D ¼ H by
G ¼ fz 2 D : jzj5 1g: ð4:18Þ
Note JG ¼ �G :¼ D � G (up to measure zero). Fundamental for us is the fact that J
essentially fixes X1ðdÞ: J:X1ðdÞ ¼ �X1ðdÞ; hence, for any z 2 D,
j0ðX1ðdÞ; zÞ ¼ j0ðJ:X1ðdÞ; zÞ ð4:19Þ
and therefore (up to measure zero)
j0ðX1ðdÞ; zÞ ¼ wGðzÞj0ðX1ðdÞ; zÞ þ J � wGðzÞj
0ðX1ðdÞ; zÞ� �
: ð4:20Þ
Here wG denotes the characteristic function of G. We haveZGnD
���Xg2G
g�j0ðX; zÞ���
¼
ZG0nD
���Xg2G
wG ðgzÞj0ðX1ðdÞ; gzÞ þ wGðJgzÞj0ðX1ðdÞ; JgzÞ
���4ZG0nD
Xg2G0
X0nG0
���Xk2Z
wGðTka gzÞj
0ðX1ðdÞ;Tka gzÞ
���þþ
ZG0nD
Xg2G0eX0 nG0
���Xk2Z
wGðJJTkb J
�1gzÞj0ðX1ðdÞ; JJTkbJ
�1gzÞ���: ð4:21Þ
310 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
Now we are in position to unfold and obtain
¼
ZG0X0
nD
���Xk2Z
wGðTka zÞj
0ðX1ðdÞ;Tka zÞ���þ
þ
ZG0eX0 nD
���Xk2Z
wGðTkb J
�1zÞj0ðX1ðdÞ;TkbJ
�1z���
¼
ZG0X0
nD
���Xk2Z
wGðTka zÞj
0ðX1ðdÞ;Tka zÞ���þ
þ
ZG00X0
nD
���Xk2Z
wGðTkb zÞj
0ðX1ðdÞ;Tkb zÞ��� ð4:22Þ
by changing variables z j�! Jz in the second integral. It is sufficient to show
the convergence of the first integral. So let F1 be a fundamental domain for
G0X0
nD and split F1 ¼ F 1 q F 2, where F 1 ¼ z 2 F1 : ImðzÞ5 1� �
. For F 1, we
have ZF 1
���Xk2Z
wGðTkazÞj
0ðX1ðdÞ;TkazÞ���
¼
ZF 1
���Xk2Z
j0ðX1ðdÞ; z þ akÞ���
¼ e�4pdZF 1
���Xk2Z
4dðx þ akÞ2
y2�
1
2p
� �e�4pd ðxþakÞ2
y2
��� dxdyy2
¼1
8d3=2ae�4pd
ZF 1
��� Xw2a�1Z
y3w2e�py2w2
e2pixw��� dxdy
y2ð4:23Þ
by Poisson summation (see Section’4.1.). But the integrand is clearly exponentially
decreasing for y ! 1. This shows the existence of this part of the integral. M More-
over, removing the absolute value signs, we see that the integral (4.23) vanishes, as
we easily conclude by interchanging the summation and the integration w.r.t. x in
the last line of (4.23). However, note that
Xk2Z
4dðx þ akÞ2
y2�
1
2p
� �e�4pd ðxþakÞ2
y2dxdy
y2ð4:24Þ
is not termwise integrable over F 1.
For F 2, we can unfold even further and getZfz2G:y4 1g
j0ðX1ðdÞ; zÞ�� �� ¼ e�4pd
Zfz2G:y4 1g
��� 4dx2
y2�
1
2p
� �e�4pdx
2
y2
��� dxdyy2
ð4:25Þ
which is clearly finite since the integrand is rapidly decreasing at the boundary of the
domain of integration. Note that the last expression does not depend on a. This
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 311
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
shows integrability for the first summand in (4.22); the second summand is handled
in the same manner! These considerations give usZGnD
Xg2G
g�j0ðX; zÞ ¼ 2
Zfz2G:y4 1g
j0ðX1ðdÞ; zÞ ð4:26Þ
¼ 2e�4pdZ
fz2G:y4 1g
4dx2
y2�
1
2p
� �e�4pdx
2
y2dxdy
y2ð4:27Þ
Using (4.13) we get
2
Zfz2G:y4 1g
j0ðX1ðdÞ; zÞ
¼ 2
Z 1
0
2
Z 1ffiffiffiffiffiffiffiffi1�y2
p e�4pd 4dx2
y2�
1
2p
� �e�4pdx
2
y2dxdy
y2
¼ 4e�4pdZ 1
0
ffiffiffiffiffiffiffiffiffiffiffiffiffi1 � y2
p2p
e�4pd1�y2
y2 y�2dy
¼1
p
Z 1
1
ffiffiffiffiffiffiffiffiffiffiffiffiw � 1
pe�4pdww�1dw ðw ¼ y�2Þ
¼1
p
Z 1
0
w12ðw þ 1Þ�1e�4pdðwþ1Þdw
¼e�4pd
2ffiffiffip
p C3
2;3
2; 4pd
� �;
where Cða; g; zÞ is the confluent hypergeometric function of the second kind which
has for ReðaÞ;ReðzÞ > 0 the integral representation
Cða; g; zÞ ¼1
GðaÞ
Z 1
0
ta�1ðt þ 1Þg�a�1e�ztdt ð4:29Þ
([16], 9.11.6). Further note the functional equation
Cða; g; zÞ ¼ z1�gCð1 þ a � g; 2 � g; zÞ ð4:30Þ
for jargðzÞj < p ([16], 9.10.8). Thus
e�4pd
2ffiffiffip
p C3
2;3
2; 4pd
� �¼
1
4ffiffiffid
ppe�4pd C 1;
1
2; 4pd
� �¼
1
4ffiffiffid
pp
Z 1
0
ðt þ 1Þ�3=2e�4pdðtþ1Þdt
¼1
4ffiffiffid
pp
Z 1
1
t�3=2e�4pdtdt: ð4:31Þ
This proves the proposition. &
312 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
4.2.3(C). The Parabolic Case
Now let ‘ be an isotropic line in VðRÞ such that its (pointwise) stabilizer G‘ is non-
trivial. Pick a point Y 2 ‘. Choose g 2 GðRÞ ¼ SL2ðRÞ such that g:Y ¼ bX0 with
X0 ¼�0 1
0 0
�and b 2 R�. Put G0 ¼ gGg�1. Hence G0
X0:¼ gGYg
�1 is equal to
f��1 ka0 1
�: k 2 Zg (if �I 2 G) for some a 2 Rþ. Recall that we defined the width of
the cusp ðY;GÞ by EðY;GÞ ¼ a=jbj ¼ EðY;GÞ.
PROPOSITION 4.8 (Theorem 3.4(iv)). Let ‘ 2 VðRÞ isotropic be a cusp of G asabove,Y 2 ‘, hY 2 QY and r 2 R�.ThenZ
GnD
XX2GðZYþhYÞ
X6¼0
j0ðrX; zÞ ¼EðY;GÞ
2pjrjð4:32Þ
where EðY;GÞ is the width of the cusp ðY;GÞ ðDef. 3:2Þ.
Proof. We can assume b ¼ �1 and by abuse of notation hY ¼ h ¼ hX0 with
h 2 ½0; 1Þ. We are interested inZGnD
XX2GðZYþhÞ
X 6¼0
j0ðX; zÞ
¼
ZGnD
Xg2GYnG
X1k¼�1
0
g�j0ðkY þ h; zÞ
¼
ZgGg�1nD
Xg2GYnG
X1k¼�1
0
j0ðg � ðkY þ hÞ; ggg�1zÞ
¼
ZG0nD
Xg2G0
X0nG0
X1k¼�1
0
g�j0ð�kX0 þ h; zÞ; ð4:33Þ
whereP01
k¼�1 omits k ¼ 0 in the case of the trivial coset. Now
j0ð�rðk þ hÞX0Þ; zÞ ¼ððk þ hÞrÞ2
y2�
1
2p
� �e�pððkþhÞrÞ2
y2dxdy
y2: ð4:34Þ
We unfold and getZG0nD
Xg2G0
X0nG0
X1k¼�1
0
g�j0ð�rðk þ hÞX0; zÞ
¼
ZG0X0
nD
X1k¼�1
0 ððk þ hÞrÞ2
y2�
1
2p
� �e�pððkþhÞrÞ2
y2dxdy
y2
¼ EðY;GÞ
Z 1
0
X1k¼�1
0 ððk þ hÞrÞ2
y2�
1
2p
� �e�pððkþhÞrÞ2
y2 y�2dy: ð4:35Þ
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 313
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
The validity of the unfolding in (4.35) (and therefore the existence of the original
integral) follows by looking at the last integral: For y ! 0 we have exponential
decay and for y ! 1 one sees-adding the constant term k ¼ 0 into the summation
if h ¼ 0–by Poisson summation (see Section 4.1.)
X1k¼�1
0 ðkrÞ2
y2�
1
2p
� �e�pðkrÞ2
y2dxdy
y2
¼1
2p� r�1y3
X1w¼�1
ðw=rÞ2e�pðyw=rÞ2 !
dxdy
y2ð4:36Þ
which is Oðy�2Þ as y ! 1. (The same argument works for h 6¼ 0.) Now in the last
expression of (4.35) interchanging summation and integration is not allowed in gen-
eral since
X1k¼�1
0 ðkrÞ2
y2e�pðkrÞ2
y2 ¼X1
k¼�1
0 1
2pe�pðkrÞ2
y2 ¼ OðyÞ ðy ! 1Þ ð4:37Þ
However, we can modify the integrand defining
FðsÞ :¼
Z 1
0
X1k¼�1
0 ððk þ hÞrÞ2
y2�
1
2p
� �e�pððkþhÞrÞ2
y2 y�2�sdy: ð4:38Þ
for s 2 C. F is holomorphic for ReðsÞ > �1 and for ReðsÞ > 0 interchanging summa-
tion and integration is valid. We obtain
FðsÞ ¼X1
k¼�1
0Z 1
0
ððk þ hÞrÞ2
y2�
1
2p
� �e�pððkþhÞrÞ2
y2 y�2�sdy ð4:39Þ
¼ 12 p
�3�s2
P1
k¼�1
0
jðk þ hÞrj�1�s�
�
Z 1
0
w �1
2
� �wðsþ1Þ=2
ewdw
ww ¼
pðk þ hÞr2
y2
� �ð4:40Þ
¼ 12 p
�3�s2 jrj�1�s zð1 þ s; hÞ þ zð1 þ s; 1 � hÞð Þ�
� Gs þ 3
2
� ��1
2G
s þ 1
2
� �� �ð4:41Þ
¼ 12 p
�3�s2 jrj�1�s zð1 þ s; hÞ þ zð1 þ s; 1 � hÞð Þ�
�s
2G
s þ 1
2
� �!
1
2pjrj; ð4:42Þ
as s ! 0. Here zðs; xÞ ¼P1
n¼0ðx þ nÞ�s is the Hurwitz zeta function which has a
simple pole of residue 1 at s ¼ 1.
314 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
5. General Signature ðp; 2Þ
5.1. THETA INTEGRAL FOR SOðp; 2Þ
Now we assume that V has signature ðp; 2Þ. Let U V be a positive definite subspace
of dimension p � 1 so that CU ¼ GUnDU is a quotient of H in M. We consider the
period integral
IjVðt;L;CUÞ ¼
ZCU
yjVðt;LÞ: ð5:1Þ
Here we write jV for j to emphasize the domain j is associated to. For p ¼ 1 and
U ¼ ð0Þ, this is the theta integral considered in the previous sections. We write
LU ¼ L \ U and LU? ¼ L \ U?: ð5:2Þ
We can split the lattice L as follows:
L ¼Xri¼1
ðli þ LUÞ ? ðmi þ LU?Þ ð5:3Þ
with li 2 L#U and mi 2 L#
U? .
THEOREM 5.1.
IjVðt;L;CUÞ ¼
Xri¼1
Wðt; li þ LUÞIjU?ðt; mi þ LU?Þ:
Here Wðt; li þ LUÞ ¼P
x2liþLUe2piqðxÞt is the standard theta function ofa cosetof the posi-
tive de¢nite latticeLU. In particular, the period integral IjVðt;L;CUÞ is a nonholomorphic
(ifCU is noncompact) modular form ofweight ðp þ 2Þ=2.Proof. By Theorem 2.1 (ii), we have that under the pullback i�U : O1;1
ðDÞ !
O1;1ðDUÞ of differential forms,
i�UjV ¼ jþU � jU? ; ð5:4Þ
where jþU is just the Gaussian on U. Then
yjVðt;LÞ ¼
Xx2L
jVðX; tÞ ð5:5Þ
¼Xri¼1
Xx2liþLU
jþUðx; tÞ
Xy2miþLU?
jU?ðy; tÞ: ð5:6Þ
integrating over CU together with the results on the theta integral for signature ð1; 2Þ
(Theorem 3.4) now gives the theorem. &
5.2. INTERSECTION NUMBERS
We will now show how one can interpret the Fourier coefficients of yjVðt;LÞ as inter-
section numbers. For a special curve C ¼ CU (dimU ¼ p � 1) and a divisor C0 ¼ CU0
(dimU0 ¼ 1), we define the intersection number in (the interior of) M:
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 315
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
½C � C0�M :¼ ½C � C0�tr
þ volðC \ C0Þ1; ð5:7Þ
the sum of the transversal intersection and the (hyperbolic) volume of the one-
dimensional intersection of C and C0. For p ¼ 2, the Hilbert modular surface case,
this follows Hirzebruch and Zagier ([7]). One easily sees that CU and CU0 intersect
transversally if and only if U þ U0 is positive definite of (maximal) dimension p, while
CU and CU0 have a one-dimensional intersection if and only if CU # CU0 , i.e., U0 # U.
THEOREM 5.2. Assume for simplicity that L ¼ LU þ LU?. Also assume that G is tor-sion-free so that M has no quotient singularities. For a composite special divisor CN
ðN 2 NÞ, we have
ðiÞ ½CU � CN�tr
¼P
N1 5 0;N2>0N1þN2¼N
rðN1;LUÞ degðN2;CUÞ;
ðiiÞ volðCU \ CNÞ1¼ rðN;LUÞvolðCUÞ:
Here
rðN;LUÞ ¼ #fx 2 LU : qðxÞ ¼ Ng
and
degðN;CUÞ ¼ #GUnfx 2 LU? : qðxÞ ¼ Ng;
the degree of the Heegnerdivisor in the modularcurveCU.Proof. We write CN ¼
Px2GnLN
Cx. We decompose x ¼ x1 þ x2 with x1 2 LU and
x2 2 LU? . We have CU Cx if and only if x2 ¼ 0. In the sum defining CN therefore
exactly the x ¼ x1 2 LU of length N contribute to the one-dimensional intersection.
This proves (ii). For (i), each x with qðx2Þ ¼ N2 > 0 contributes. Taking into account
the action of GU on LU? gives the assertion. &
We certainly have a similar theorem if L does not split along U. If G is not torsion-
free, then we can always pass to a torsion-free subgroup G0 of finite index to obtain
intersection numbers on G0nD. The intersection numbers on GnD (in the sense of
rational homology manifolds) are then obtained by dividing by the degree of the
covering G0nD j�!GnD.
COROLLARY 5.3. Write
IjVðt;L;CUÞ ¼
X1N¼0
cðNÞqN þX1
N¼�1
cðN; vÞqN
for the Fourier expansion of the above theta integral.Then
cðNÞ ¼ ½CU � CN�M;
i.e., the intersection numbers are exactly the Fouriercoe⁄cients of the holomorphic partofIjV
ðt;L;CUÞ.Proof. Just write down the Fourier expansion of IjV
ðt;L;CUÞ using Theorem 3.5
and 5.1. &
316 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
5.3. HIRZEBRUCH–ZAGIER CASE
As an example we will derive the basic case of the results of Hirzebruch and
Zagier [7] and its (mild) extensions by Franke [5], Hausmann [6] and van der
Geer [23].
Let D > 0 be the discriminant of the real quadratic field K ¼ QðffiffiffiffiD
pÞ over Q, OK
its ring of integers. We denote by x j�!x0 the Galois involution on K. We let
V M2ðKÞ be the space of skew-Hermitian matrices in M2ðKÞ, i.e., which satisfy
the relation tX0 ¼ �X. V is a vector space over Q of dimension 4. We let L V
be the integral skew-Hermitian matrices; that is
L ¼ X ¼affiffiffiffiD
pl
�l0 bffiffiffiffiD
p
� �: a; b 2 Z; l 2 OK
: ð5:8Þ
As usual, the determinant defines a quadratic form on L; we have QðXÞ ¼ abD þ ll0,
which is of signature ð2; 2Þ and has Q-rank 1, i.e., it splits over Q into a hyperbolic
plane and into an anisotropic part of rank 2. SL2ðOKÞ acts on L by g � X ¼ gXtg0 asisometries. We let G be the Hurwitz–Maass extension of SL2ðOKÞ which is the maxi-
mal discrete subgroup of PGLþ2 ðRÞ
2 containing SL2ðOKÞ (for a brief discussion of its
definition and properties, see [23]; for example, if D � 1mod 4 is a prime, then
G ¼ SL2ðOKÞ . This is actually the case Hirzebruch and Zagier considered). Slightly
changing our notation we write S ¼ GnD for the Hilbert modular surface. The
Hirzebruch–Zagier cycles TN are nothing but our special cycles CN ¼P
X2LNCX.
TN has finitely many components, is nonempty if ðN=pÞ 6¼ �1 and is compact if N
is not the norm of an ideal in OK. We can compactify S by adding the cusps and
resolving the singularities thus created. We call this (up to quotient singularities)
nonsingular compact surface ~S. Now H2ð ~SÞ decomposes canonically into as the
direct sum of the image of H2ðSÞ and the subspace generated by the homology cycles
of the curves of the cusp resolution. We denote by TcN the component in the first fac-
tor of TN. Hirzebruch and Zagier compute the intersection numbers ½TcN � TM�
(N;M 2 N) and by a direct computation they show that these numbers are the Four-
ier coefficients of a holomorphic modular form of weight 2. In fact, ½TcN � TM� is the
sum of the intersection numbers ½TN � TM�S in the interior and the contribution of
the cusp resolution, and these numbers individually occur as Fourier coefficients
of two nonholomorphic modular forms. For M ¼ 1, one has
ðT1 � TNÞS ¼ HDðNÞ ¼X
s244Ns2�4NmodD
H4N � s2
D
� �; ð5:9Þ
where HðNÞ denotes the class number of positive definite binary quadratic forms, see
Example 3.9. We now recover these results: We consider the curve T1 ¼ C1 in our
setting. G acts transitively on L1, the vectors of length 1, see [6,23]. So T1 ¼ CX0
for any X0 2 L1. We pick X0 ¼�
0 1
�1 0
�2 L.
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 317
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
THEOREM 5.4.
1
2
ZT1
yjVðt;LÞ ¼
X1N¼0
HDðNÞqN þ2v�1=2ffiffiffiffi
Dp
Xl2O
b 4pl � l0
2
� �2
v
!qll
0
:
So the Fourier coe⁄cients of the holomorphic part of the period integralRT1yjV
ðt;LÞ arethe intersection numbers of the cyclesT1 andTN in the‘interior’ofGnD.
Proof. First we assume D � 0mod 4, so K ¼ Qðffiffiffid
pÞ with d ¼ D=4 square free. In
this case L splits orthogonally:
L ¼ ZX0 ? ðL \ X?0 Þ ð5:10Þ
and
LU? ¼ L \ X?0 ¼
ffiffiffid
p 2a b�b 2c
� �: a; b; c 2 Z
; ð5:11Þ
which is—up to the scaling—exactly the lattice considered in Example 3.9(i) which
gave rise to Zagier’s Eisenstein series F !. Hence by Theorem 5.1 and Example
3.9(i) we find
12Ijðt;L;T1Þ ¼ WðtÞIjU?
ðt;LU?Þ ð5:12Þ
¼ WðtÞF ðdtÞ ð5:13Þ
¼Xm2Z
qm2
!�
�X1N¼0
HðNÞqdN þ ðdvÞ�1=2Xn2Z
bð4pn2dvÞ q�dn2
!ð5:14Þ
¼Xs2 4N
s2�NðdÞ
HN � s2
d
� �qNþ
þv�1=2ffiffiffi
dp
X1N¼�1
Xn;m2Z
m2�dn2¼N
bð4pdn2vÞ qN ð5:15Þ
¼X
ð2sÞ2 4 4N
ð2sÞ2�4N ð4dÞ
H4N � ð2sÞ2
4d
� �qNþ
þ2v�1=2ffiffiffiffiffi
4dp
Xl2O
b 4pl � l0
2
� �2
v
!qll
0
ð5:16Þ
318 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
since O ¼ fm þ nffiffiffid
p: m; n 2 Zg. Noting D ¼ 4d, the theorem follows for
D � 0mod 4. The case D � 1mod 4 is slightly more complicated since L does not
split orthogonally so that one has to deal with cosets. We have
LU? ¼ L \ X?0 ¼
ffiffiffiffiD
p a b�b c
� �: a; b; c 2 Z
ð5:17Þ
which is the lattice from Example 3.9(ii)! We set X1 ¼�
0ffiffiffiD
pffiffiffiD
p0
�and obtain
L ¼ ZX0 ? LU?ð Þ $ 12ðX0 þ X1Þ þ ZX0 ? LU?ð Þ: ð5:18Þ
The holomorphic part is given byX1j¼0
Xm2Z
q mþj2ð Þ
2
! X1n¼0
Hð4n � jÞqD n� j4ð Þ
!¼X1j¼0
X1N¼0
Xm;n2Z
�Hð4n � jÞqN; ð5:19Þ
where the inner summation extends over all integers m and n such that
m þj
2
� �2
þD n �j
4
� �¼ N
or, equivalently,
4n � j ¼4N � ð2m þ jÞ2
D:
So (5.19) becomesX1j¼0
X1N¼0
Xð2mþjÞ2 4 4N
ð2mþjÞ2�4N ðDÞ
H4N � ð2m þ jÞ2
D
� �qN
¼X1N¼0
Xs2 4 4N
s2�4N ðDÞ
H4N � s2
D
� �qN: ð5:20Þ
For the nonholomorphic part, we first note
O ¼ m þ n1 þ
ffiffiffiffiD
p
2: m; n 2 Z
:
For l ¼ m þ n1 þ
ffiffiffiffiD
p
2we write l % ðm; nÞ. Then
X1j¼0
Xm2Z
q mþj2ð Þ
2
!2ðvDÞ
�1=2Xn2Z
b 4p n þj
2
� �2
vD
!q�D nþ j
2ð Þ2
!ð5:21Þ
¼2v�1=2ffiffiffiffi
Dp
X1j¼0
X1N¼0
Xm;n2Z
�b 4p n þj
2
� �2
Dv
!qN; ð5:22Þ
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 319
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
where the inner sum goes over all m; n such that ðm þ ðj=2ÞÞ2 � Dðn � ðj=4ÞÞ ¼ N.
¼2v�1=2ffiffiffiffi
Dp
X1j¼0
XN2Z
Xl%ðn�m;2nþjÞ
ll0¼N
b 4pl � l0
2
� �2
v
!qN ð5:23Þ
¼2v�1=2ffiffiffiffi
Dp
Xl2O
b 4pl � l0
2
� �2
v
!qll
0
; ð5:24Þ
as desired. &
The factor 12 in the previous theorem occurs as �1 2 G, which acts trivially on
D ’ H � H. To obtain the intersection numbers ½Tc1:TN� and the holomorphic modu-
lar form of weight 2, one can now apply the holomorphic projection principle ontoRT1yjV
ðt;LÞ. This is an idea of van der Geer and (independently) Zagier, which has
been carried out in [23].
Remark 5.5. We see that Theorem 5.1 is the exact generalization of a part of the
results of Hirzebruch–Zagier. One is certainly very interested to obtain complete
analogues of these results. Applying holomorphic projection onto the functionRCU
yjVðt;LÞ in Theorem 5.1 for p > 2 is certainly possible, but this does not have a
priori a geometric interpretation. More promising seems to be an analysis of the
boundary along the lines of the theory of Kudla and Millson. It seems likely that
Eisenstein cohomology will enter the picture at this stage. We hope to come back to
this issue.
Acknowledgements
Most of the results in this paper are part of the author’s PhD thesis at the University
of Maryland. I would like to thank my advisor Steve Kudla for his guidance
throughout the last years. I also would like to thank John Millson for many valuable
discussions. The final stages of this paper were completed while visiting the Max-
Planck-Institut f �ur Mathematik in Bonn as a fellow of the European PostDoc Insti-
tute (EPDI). I would like to thank its directors in general for providing such a stimu-
lating environment and in particular Professor Zagier for some comments on the
results of this paper.
References
1. Borcherds, R.: Automorphic forms with singularities on Grassmannians, Invent. Math.132 (1998), 491–562.
2. Borcherds, R.: The Gross–Kohnen–Zagier theorem in higher dimensions, DukeMath. J.97 (1999), 219–233.
3. Bruinier, J.: Borcherds products and Chern classes of Hirzebruch–Zagier divisors, Invent.
Math. 138 (1999), 51–83.
320 JENS FUNKE
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at
4. Bruinier, J.: Borcherds products on Oð2; lÞ and Chern classes of Heegner divisors, Habi-
litationsschrift, Universit�at Heidelberg (2000).5. Franke, H.-G.: Kurven in Hilbertschen Modul£ �achen und Humbertsche Fl �achen, Bonner
Math. Schriften, 114, Univ. Bonn, 1978.
6. Hausmann, W.: Kurven auf Hilbertschen Modul£ �achen, Bonner Math. Schriften, 123,Univ. Bonn, 1980.
7. Hirzebruch, F. and Zagier, D.: Intersection numbers of curves on Hilbert modular sur-
faces and modular forms of Nebentypus, Invent. Math. 36 (1976), 57–113.8. Kudla, S.: On the integrals of certain singular theta-functions, J. Fac. Sci.Univ. Tokyo 28
(1982), 439–465.9. Kudla, S.: Algebraic cycles on Shimura varieties of orthogonal type, Duke Math. J. 86
(1997), 39–78.10. Kudla, S.: Central derivatives of Eisenstein series and height pairings, Ann. of Math. 146
(1997), 545–646.
11. Kudla, S.: Derivatives of Eisenstein series and generating functions for arithmetic cycles,Se¤ minaire Bourbaki 876 (2000).
12. Kudla, S. and Millson, J.: The theta correspondence and harmonic forms I, Math. Ann.274 (1986), 353–378.
13. Kudla, S. and Millson, J.: The theta correspondence and harmonic forms II, Math. Ann.277 (1987), 267–314.
14. Kudla, S. and Millson, J.: Intersection numbers of cycles on locally symmetric spaces andFourier coe⁄cients of holomorphic modular forms in several complex variables, IHESPubl. 71 (1990), 121–172.
15. Kudla, S. Rapoport, M. and Yang, T.: On the derivative of an Eisenstein series of weightone, Internat Math. Res. Notices 7 (1999), 347–385.
16. Lebedev, N. N.: Special Functions and their Applications, Dover, New York, 1972.17. Lion, G. and Vergne, M.: The Weil Representation, Maslov Index and Theta Series,
Progr. in Math. 6, Birkh�auser, Basel, 1980.18. Miyake, T.: Modular Forms, Springer, New York, 1989.19. Oda, T.: On modular forms associated with inde¢nite quadratic forms of signature
ð2; n � 2Þ, Math. Ann. 231 (1977), 97–144.20. Rallis, S. and Schiffmann, G.: On a relation between gSL2 cusp forms and cusp forms on
tube domains associated to orthogonal groups, Trans. Amer. Math. Soc. 263(1) (1981),1–58.
21. Sato, F.: On zeta functions of ternary zero forms, J. Fac. Sci. Uni. Tokyo 28 (1982), 585–604.
22. Shintani, T.: On zeta functions associated with the vector space of quadratic forms, Sci.Uni. Tokyo 22 (1975), 25–65.
23. van der Geer, G.: Hilbert Modular Surfaces, Ergeb. Math. Grenzgeb. (3) 16, Springer,New York, 1988.
24. Wang, S. P.: Correspondence of modular forms to cycles associated toOðp; qÞ, J. Di¡er-ential Geom. 18 (1985), 151–223.
25. Zagier, D.: Nombres de classes et formes modulaires de poids 3=2, C.R. Acad. Sci. ParisSe¤ r. A^B 281 (1975), 883–886.
HEEGNER DIVISORS AND NONHOLOMORPHIC MODULAR FORMS 321
https://www.cambridge.org/core/terms. https://doi.org/10.1023/A:1020002121978Downloaded from https://www.cambridge.org/core. IP address: 54.39.106.173, on 11 Sep 2020 at 03:46:36, subject to the Cambridge Core terms of use, available at